Engineered NK Cells Target CD73 to Attack Hard-to-Treat Lung Adenocarcinoma

Relapsed or refractory lung adenocarcinoma remains one of the hardest problems in thoracic oncology. Once a tumor has returned after initial treatment or stopped responding to standard therapies, options narrow sharply and prognosis worsens. A new preclinical study now reports that immune cells engineered to seek out a molecule called CD73 can kill such tumors in laboratory dishes and in several mouse models, including mice carrying tumor fragments from patients whose disease had relapsed or become resistant to treatment. The work, published in Biomarker Research, offers a detailed proof of concept rather than a clinical result, but it maps out a strategy that combines two attacks in one: destroying cells that display CD73 and neutralizing the immunosuppressive signal that CD73 helps generate.

CD73, encoded by the NT5E gene, is an enzyme sitting on the surface of many cells. Its rate-limiting role in producing extracellular adenosine makes it a double-edged molecule in cancer biology. Adenosine accumulating in the tumor neighborhood dampens the activity of T cells and natural killer cells, the immune soldiers that would otherwise recognize and destroy malignant cells. Beyond this enzymatic function, CD73 also appears to promote epithelial-to-mesenchymal transition and metastatic spread through mechanisms that do not depend on adenosine at all. Among solid tumors, non-small cell lung cancer shows some of the highest CD73 expression, a pattern further amplified by common oncogenic drivers such as mutant EGFR and KRAS. Antibodies and small molecules designed to block CD73 are already in clinical testing, but these approaches lack direct tumor-killing capacity and cannot address the non-enzymatic contributions of the protein.

The research team, working across institutions in China, first asked whether CD73 is actually enriched in relapsed or refractory disease. Using public data from the TCGA lung adenocarcinoma cohort, they noted that the two recurrent tumor samples available showed higher NT5E mRNA expression than the distributions seen in normal lung and primary tumors. The authors themselves flagged this as descriptive rather than biomarker validation, given the tiny sample size. Their own immunohistochemical cohort, though also small, compared three primary tumors with six relapsed or refractory specimens and found higher CD73 protein in the latter group. Kaplan-Meier analysis using Human Protein Atlas data showed that patients whose tumors expressed high CD73 had significantly worse overall survival than those with low expression, an association rather than a causal claim.

With this target profile in hand, the investigators built a second-generation chimeric antigen receptor directed against CD73. The construct paired an anti-CD73 single-chain variable fragment with a 4-1BB costimulatory domain and a CD3ζ signaling module, the same architectural logic used in many CAR designs. Lentiviral delivery introduced the receptor into NK-92 cells, an immortalized natural killer cell line, achieving roughly fifty percent transduction efficiency before antibiotic selection expanded the positive population. Bulk RNA sequencing then compared engineered and parental cells, revealing differential expression of genes tied to cytokine signaling, antigen processing, oxidative phosphorylation, TNF signaling, and ferroptosis. Selected cytotoxicity genes such as FASLG, GZMA, and GZMB rose, while inhibitory receptors including KLRC1, KLRD1, and HAVCR2 fell. The authors emphasized that these changes were selective rather than a uniform activation program, suggesting targeted remodeling of the effector phenotype.

Functional testing confirmed that the engineered cells behaved as intended. When co-cultured with H1299 cells, the CD73-CAR-NK-92 group released markedly more perforin, granzyme B, TNF-α, and IFN-γ than parental NK-92 cells. Flow cytometry showed upregulation of the degranulation marker CD107a along with FASL, NKG2D, and NKp46, while the inhibitory receptors TIGIT and NKG2A declined. Cytotoxicity assays across a graded panel of targets told a consistent story: killing of CD73-low A549 cells stayed comparable to unmodified NK-92 cells, whereas engineered cells achieved near-complete destruction of CD73-high PC-9 and H1299 cells, where parental cells managed roughly fifty to sixty percent. Neither cell type showed appreciable toxicity toward BEAS-2B cells, with killing remaining below twenty percent across all effector-to-target ratios tested. Genetic perturbations sealed the antigen-dependence argument, as siRNA knockdown of CD73 in H1299 cells eroded the CAR advantage and enforced CD73 overexpression in A549 cells boosted CAR-mediated killing.

Perhaps the most mechanistically interesting finding concerns adenosine. Because CD73 generates the adenosine that suppresses NK function through A2A receptor signaling, a CAR targeting CD73 might do more than kill; it might also disarm the suppressive circuit. The researchers measured phosphorylation of PKA substrates, a downstream readout of A2AR-cAMP-PKA signaling, in NK cells exposed to exogenous adenosine or to H1299 co-culture. Unmodified NK-92 and peripheral blood NK cells showed increased phosphorylation under both conditions, while CD73-engineered versions did not. Exogenous adenosine at ten micromolar markedly impaired parental NK-92 cytotoxicity, yet the engineered cells largely retained killing activity across the tested ratios. Engineered cells also lowered extracellular adenosine in co-culture, likely by eliminating the CD73-high tumor cells that serve as a major adenosine source. A2A receptor blockade partially rescued parental NK-92 function, and an anti-CD73 antibody similarly enhanced killing, framing the CAR approach as one that both removes the adenosine factory and resists the adenosine signal itself.

To test whether this behavior generalizes beyond the NK-92 platform, the team engineered peripheral blood-derived NK cells, umbilical cord blood-derived NK cells, and iPSC-derived NK cells with the same receptor. Enhanced cytotoxicity against PC-9 and H1299 cells, but not A549, appeared consistently across all three sources in both two-dimensional real-time assays and three-dimensional spheroid models. This reproducibility matters for translational ambitions, since allogeneic off-the-shelf CAR-NK products would likely draw from these more clinically practical sources rather than the irradiated NK-92 line used in most of the in vivo work.

In mouse models, the story grew more compelling. Subcutaneous H1299 xenografts in NCG mice progressed rapidly in the PBS group, with all animals reaching ethical endpoints by day twenty-one. Parental NK-92 treatment modestly delayed growth, leaving a single survivor, while CD73-CAR-NK-92 treatment suppressed progression enough that all mice remained alive at that timepoint. Flow cytometry on days three and seven showed greater abundance of engineered cells in peripheral blood and greater accumulation within tumors, consistent with improved persistence and trafficking. Intratumoral levels of granzyme B, TNF-α, and IFN-γ rose in the CAR group. A parallel PC-9 xenograft model reproduced the pattern. In an experimental lung metastasis model using intravenous H1299-luciferase inoculation, bioluminescence signals diverged sharply from day ten onward, with the CAR group showing sustained decline and near-complete clearance in some animals, while PBS and NK-92 groups climbed. Gross examination and histology at day seventeen confirmed fewer and smaller metastatic lesions in the CAR-treated lungs.

The most clinically relevant experiments used patient-derived xenografts. In a conventional PDX model built from a relapsed or refractory patient sample, CD73-CAR-NK-92 treatment produced marked tumor regression where PBS allowed continued growth and NK-92 only modestly delayed progression. Splenic NK-cell proportions rose roughly eightfold compared with the NK-92 group, and activated CD107a-positive NK cells increased approximately tenfold in spleen and two-and-a-half-fold in tumors. Intratumoral perforin, granulysin, TNF-α, and IFN-γ all climbed, while adenosine dropped within tumors but stayed unchanged in peripheral blood, pointing to compartmentalized rather than systemic immune activation. An immune-humanized PDX model, which adds a human immune system to mice bearing patient tumor fragments, extended these findings. Engineered cells drove rapid tumor dissipation, with complete clearance in several mice, and flow cytometry revealed increased infiltration and activation of NK cells, dendritic cells, CD8-positive T cells, and CD4-positive T cells within tumors. Bulk RNA sequencing of the engineered cells identified upregulated CCL3, CSF1, CSF2, CXCR3, CXCR6, and ACKR3, and conditioned medium from these cultures promoted T-cell and dendritic-cell migration in Transwell assays, an effect significantly reduced by CCL3 neutralization. Serum IL-6, CRP, and hepatorenal markers showed no treatment-related deterioration over a twenty-day observation window, and organ histology revealed no overt abnormalities.

The authors are candid about the limits of this evidence. Clinical NK-92 products require irradiation before infusion, which may alter persistence and durability compared with the non-irradiated cells used here. Systemic IL-2 support was given without an IL-2-only control, complicating comparisons with the PBS group, and the need for exogenous IL-2 itself raises translatability concerns given its known toxicity. The clinical cohort was small, the safety data rest on limited animal numbers and short follow-up, and potential on-target off-tumor toxicity from CD73 expression in normal tissues remains unassessed. Conventional randomization, stratification, and blinding were not implemented in the PDX cohorts, and human immune reconstitution relied on supplier batch-level quality control rather than longitudinal monitoring. Whether CD73 antigen loss emerges under sustained CAR pressure, a recognized resistance mechanism in solid tumor CAR-T therapy, awaits investigation. Future work, the team suggests, should test peripheral blood-derived or iPSC-derived engineered cells in clinically relevant models and explore combinations with PD-1 or PD-L1 inhibitors. For now, the study stands as a carefully layered preclinical argument that CD73 is a rational target in relapsed or refractory lung adenocarcinoma and that a CAR-NK strategy aimed at it merits progression toward early-phase clinical evaluation.

Subject of Research: Preclinical evaluation of CD73-targeted CAR-NK cell therapy for relapsed or refractory lung adenocarcinoma

Article Title: CD73-targeted CAR-NK cell therapy displays potent antitumor activity and translational potential in relapsed/refractory lung adenocarcinoma

Article References: Liu, Q., Dong, Q., Chen, Y., Wang, N., Zhuang, D., Chen, L., Ye, Y., Huang, S., Zhong, W., Zhang, W. K., Xu, T., Sun, J., & Liu, H. (2026). CD73-targeted CAR-NK cell therapy displays potent antitumor activity and translational potential in relapsed/refractory lung adenocarcinoma. Biomarker Research, 14(1), Article 154. https://doi.org/10.1186/s40364-026-01036-5

Image Credits: AI Generated

DOI: 10.1186/s40364-026-01036-5

Keywords: CAR-NK cells, CD73, lung adenocarcinoma, relapsed refractory cancer, adenosine immunosuppression, natural killer cells, patient-derived xenograft, immune-humanized mice, preclinical study, tumor microenvironment, immunotherapy, NT5E